Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Evaluation Metrics for Automated Typographic Poster Generation

Sérgio M. Rebelo, J. J. Merelo, João Bicker, Penousal Machado

TL;DR

The paper addresses the challenge of objectively evaluating computational typographic poster designs by proposing a ten-metric framework that covers legibility, aesthetics, and semantics. It couples these metrics with a constrained evolutionary algorithm to generate posters from varying texts, leveraging emotion recognition to emphasize semantic content. Key contributions include the metric set itself, a constrained evolution workflow, and an open-source implementation, demonstrated across multilingual inputs. The work advances design automation by enabling systematic, reproducible evaluation and optimization of typographic posters, with potential for integration into autonomous design pipelines.

Abstract

Computational Design approaches facilitate the generation of typographic design, but evaluating these designs remains a challenging task. In this paper, we propose a set of heuristic metrics for typographic design evaluation, focusing on their legibility, which assesses the text visibility, aesthetics, which evaluates the visual quality of the design, and semantic features, which estimate how effectively the design conveys the content semantics. We experiment with a constrained evolutionary approach for generating typographic posters, incorporating the proposed evaluation metrics with varied setups, and treating the legibility metrics as constraints. We also integrate emotion recognition to identify text semantics automatically and analyse the performance of the approach and the visual characteristics outputs.

Evaluation Metrics for Automated Typographic Poster Generation

TL;DR

The paper addresses the challenge of objectively evaluating computational typographic poster designs by proposing a ten-metric framework that covers legibility, aesthetics, and semantics. It couples these metrics with a constrained evolutionary algorithm to generate posters from varying texts, leveraging emotion recognition to emphasize semantic content. Key contributions include the metric set itself, a constrained evolution workflow, and an open-source implementation, demonstrated across multilingual inputs. The work advances design automation by enabling systematic, reproducible evaluation and optimization of typographic posters, with potential for integration into autonomous design pipelines.

Abstract

Computational Design approaches facilitate the generation of typographic design, but evaluating these designs remains a challenging task. In this paper, we propose a set of heuristic metrics for typographic design evaluation, focusing on their legibility, which assesses the text visibility, aesthetics, which evaluates the visual quality of the design, and semantic features, which estimate how effectively the design conveys the content semantics. We experiment with a constrained evolutionary approach for generating typographic posters, incorporating the proposed evaluation metrics with varied setups, and treating the legibility metrics as constraints. We also integrate emotion recognition to identify text semantics automatically and analyse the performance of the approach and the visual characteristics outputs.
Paper Structure (7 sections, 3 equations, 5 figures, 2 tables)

This paper contains 7 sections, 3 equations, 5 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Metrics
  4. Experiments
  5. Experimental Setup
  6. Experimental Results
  7. Conclusions and Future Work

Figures (5)

  • Figure 1: Progression of constraint penalty and legibility metrics across generations. Visualised results present the average of three stages, totalling 1,800 runs. Black lines depict penalty constraint values, while pink and blue lines show the scores of text legibility and grid appropriateness metrics, respectively. Fittest individuals are depicted with solid lines, and the average population with shaded lines.
  • Figure 2: Evolutionary progress in the S1 experiment. The results are averages from 600 runs, considering varied text inputs. The chart on the left illustrates best individual fitness (solid lines), constraint penalty (dashed lines) and maximum fitness (dotted). On the right, the chart shows score metrics related to the significance of semantics in layout (green lines) and typography (orange lines). Solid lines represent the average of the fittest individuals, while shaded lines represent the average of the entire population.
  • Figure 3: Examples of designs evolved during the three experimental stages. More results can be found in the supplementary material folder.
  • Figure 4: Evolutionary progress in the S2 experiment. The results are averages from 600 runs, considering multiple text inputs. The chart on the left illustrates best individual fitness (solid lines), constraint penalty (dashed lines) and maximum fitness (dotted). On the right, the chart shows score metrics related to aesthetics, including alignment (navy), regularity (green), justification (violet), typography pairing (pink), balance (turquoise), and negative space fraction (orange). Solid lines represent the average of the fittest individuals, while shaded lines represent the average of the entire population.
  • Figure 5: Progression of evolution in the S3 experiment. The results are averages from 600 runs, considering various text inputs. The top-left chart presents the progression of the best individual fitness (solid lines), constraint penalty (dashed lines) and maximum fitness (dotted). The top-right and bottom charts illustrated the scores of metrics related to semantics and aesthetics, respectively. Semantic metrics include the significance of semantics in layout (green) and typography (orange). Aesthetic metrics include alignment (navy), regularity (green), justification (violet), typography pairing (pink), balance (turquoise), and negative space fraction (orange). Solid lines represent the average of the fittest individuals, while shaded lines represent the average of the entire population.